Structural basis of Tie2 activation and Tie2/Tie1 heterodimerization

Structure of Tie2 Fn-Like Domains 1–3.

To better understand the mechanism of Tie2 dimerization and activation, we expressed Tie2 Fn-like domains 1–3 (Fn1–3) (Fig. 1A) in insect cells and crystallized the purified protein in space groups C2 and P2₁. The crystal structure in space group C2 was determined at 2.9-Å resolution using multiple isomorphous replacement with anomalous scattering (MIRAS) phases, and the structure in space group P2₁ was determined at 2.6-Å resolution with molecular replacement using the structure in the C2 space group as a search model (Table S1). Both structures revealed a homodimer of the Tie2 Fn-like domains in the asymmetric unit (Fig. 1B). The dimerization is based on symmetrical interfaces between the membrane-proximal Fn3 domains, placing the C termini close to each other at a distance of about 25 Å. The C termini point in the same direction, whereas the N-terminal Fn-like domains are directed away from each other. Dimerization is mediated mainly by hydrogen bonds between the main-chain atoms of antiparallel β-strands in residues Asp682 to Lys690. Other interactions occur between symmetry-related Lys700 and Gly701, and the interacting residues are largely conserved (Fig. 1C). According to the Pisa server analysis, the interface buries a surface area of about 580 Ų per chain in space group C2; the solvation free energy gains of −1.5 and −1.9 kcal/mol for each chain and the related P value of 0.54 suggest considerable specificity (Tables S2 and S3). In space group P2₁, the interactions between the antiparallel β-strands are almost identical, but, because of additional interactions, the buried surface area is larger, about 700 Ų per chain (Tables S2 and S3).

The Tie2 homodimers in the two crystal structures share the same overall structure, except that only one of the two Tie2 proteins in the C2 dimer has the Fn1 domain (Fig. S1). The 376 Cα atoms in the two Tie2 Fn2–3 dimers superimpose with an rmsd of 1.445 Å. The crystal packing indicates that the Fn1 domains were proteolytically removed before crystallization in space group P2₁, whereas in space group C2 the other Fn1 domain is either disordered or present at low occupancy, because crystal packing in C2 should accommodate it, but only weak residual density is seen (Fig. S2). Superposition of the Fn1 domain-bearing chain in space group C2 with the chain lacking the Fn1 domain creates a model of a Tie2 Fn1–3 homodimer (Fig. 1D). A recent EM analysis of full-length Tie2 ECDs did not indicate Tie2 dimerization, but the shape and length of the extended structure of Tie2 Fn-like domains 1–3 in space group C2 are consistent with those in the EM analysis (11.8 nm vs. 12 nm) (37). Although data from small-angle X-ray scattering (SAXS) of an Fn2–3 protein (at ∼80 μM) indicated a structure that fits best with our crystal structure of an Fn3 domain-mediated Tie2 dimer, the Tie2 ECD and Tie2 Fn domains were monomeric in solution when analyzed by size-exclusion chromatography multiangle laser light scattering (SEC-MALLS) (Fig. S3 A–E). This finding suggests that the unliganded receptor has a relatively weak tendency for the Fn-domain–mediated interactions and that the formation of the putative homotypic interactions by the full-length receptor requires spatial (and dimensional) constraint in the plasma membrane and/or is ligand dependent.

Homotypic Fn3 Interactions Are Required for Ligand-Induced Tie2 Activation.

We tested the model of Fn3-mediated homotypic interactions of Tie2 by targeting the dimerization interface with a V685Y/V687Y double mutation and by expressing the WT and mutant receptors in HeLa cells for ligand-stimulation experiments. Two conserved valine residues were changed to bulkier tyrosine residues to introduce steric hindrance across the Fn3–Fn3 interface (Fig. 2 A and B). A model of the Tie2 Fn3 mutant indicates that the neighboring Tyr674 and Tyr697 residues force the tyrosine mutant conformations toward the Fn3–Fn3 interface (Fig. 2B). Although Comp-Ang1 stimulation of the WT Tie2 induced strong Tie2 phosphorylation and colocalization with Tie1 in cell–cell junctions of receptor-transfected HeLa cells, phosphorylation of the mutant Tie2 was much reduced, strongly suggesting that the mutation interfered with Tie2 dimerization (Fig. 2 C and D) (10, 18). Neither these Fn3 mutations nor the other Tie2 or Tie1 mutations described below had any overt effect on receptor folding or expression in HeLa cells (Fig. S3 F and G). The Fn1–Fn2 junction in Tie2, unlike the Fn2–Fn3 junction, has no specific interactions between the two domains, consistent with its apparent flexibility and susceptibility to proteolysis (Fig. S4 A–C). In comparison with the analysis of the Fn1 domain of the Ang1–Tie2 LBD complex at 4.5-Å resolution (33), we observed a sequence register difference of one to three residues between the Fn1 structures (Fig. S4D). However, by superimposing the Fn1 domains in the Tie2 Fn1–3 homodimer and Ang1–Tie2 LBD complex structures (33), we were able to construct a structural model of the Ang1-bound Tie2 ECD homodimer that provided mechanistic insight for Tie2 activation in cis (Fig. 3 A and B). Interestingly, the Tie2 LBDs are too far apart for a dimeric angiopoietin ligand to bridge the receptors for successful dimerization and activation. Instead, a dimeric ligand would compete with multimeric angiopoietin for Tie2 binding (Fig. 3B).

A Model of Tie1/Tie2 Heterodimerization.

To study the mechanism of Tie1 and Tie2 heterodimerization, we expressed the membrane-proximal Tie1 Fn3 in insect cells and crystallized the purified protein for structure determination. The structure was solved at 2.6-Å resolution by molecular replacement using the Tie2 Fn3 domain as a search model (Table S1). Interestingly, the structure is a strand-swapped homodimer in which the two Fn-like domains exchange their C-terminal β-strands (Fig. S5 A and B). Strand swapping occurs by extension of the sixth β-strand toward its counterpart to form an antiparallel β-sheet that extends over Ile722–Gly724 in the bridge region where the two strands meet. This swapping may be an important feature of Tie1 function, but a model derived from SAXS data indicates that Tie1 Fn3 is a monomer rather than a dimer in solution (Fig. S5C). Thus, the strand-swapped dimerization may instead be a crystallization artifact of an isolated domain.

Comparison of the Tie1 and Tie2 Fn3 domain structures reveals very similar folds in both, despite their low sequence identity (Fig. 4A), suggesting that structural homology between the Tie receptors extends to the Fn3-mediated dimerization and activation. Superposition of Tie1 Fn3 domain with an Fn3 domain in the Tie2 homodimer indicates that Tie1/Tie2 heterodimerization also could be mediated by interactions between the antiparallel β-strands in the Fn3 domains (Fig. 4B). The model also suggests additional favorable interactions, including a salt bridge between Tie2 Asp682 and Tie1 Arg697 (Fig. 4B). We tested the model of Fn3-mediated Tie2/Tie1 heterodimerization by analyzing Tie1 phosphorylation in Tie1/Tie2-transfected HeLa cells. We created a Tie1 tyrosine mutant (3Y; V681Y, L691Y, and I693Y) in an analogous manner to the Tie2 Fn3 double mutant. Val681 was mutated to tyrosine to mimic the Tie2 Fn3 domain and to direct the side-chain conformers of I693Y and L695Y toward the interface. Another mutant (5M) with additional D689R and D696R substitutions was designed to disrupt the interactions further. The Tie1 mutants increased the basal phosphorylation of Tie1 and made Tie1 insensitive to Tie2-mediated Comp-Ang1 stimulation (Fig. 4 C and D). Interestingly, the 5M mutant did not affect Tie1/Tie2 heterodimerization according to the cell-surface crosslinking assay, and Tie1 baseline phosphorylation was not affected in Tie1 singly transfected cells, suggesting that the Tie1 Fn3 mutations affected only Tie1/Tie2 kinase crosstalk (Fig. S6 A–C).

A Model for Tie2 Clustering.

In addition to the very similar interactions between the Fn3 domains in the C2 and P2₁ structures, we observed structurally analogous interactions between the Fn2 domains of neighboring homodimers in the crystal, suggesting that both types of interactions exhibit natural protein contacts (Fig. 5A). Combining Fn3-guided homodimerization of Tie2 with Fn2-guided interactions between homodimers creates an array of Tie2 molecules that could represent the structural basis of Tie2 clustering (Fig. S7A). The symmetrical interactions between the Fn2 domains involve multiple hydrogen bonds between highly conserved Arg-Trp-Arg motifs and the main-chain atoms of residues 582–585 in a hairpin loop of the neighboring chain (Fig. S7 B and C). Interestingly, in a primary congenital glaucoma (PCG) family, a TEK mutation (Y611C) is located close to this Fn2′–Fn2′′ interface of neighboring homodimers, resulting in haploinsufficiency because of the loss of Tie2 function (Fig. 5B) (38). Tyr611 is a buried residue that packs against the hairpin loop, making interactions with the Arg-Trp-Arg motif. Mutation of Tyr611 to cysteine would remove a potentially important hydrogen bond to His606 in the neighboring Fn2 loop. We next created a Q588R/V612R double mutant to target the Fn2′-Fn2′′ interface and analyzed the effect of the Fn2 mutations on Tie2 activation in Tie2 singly transfected and Tie1/Tie2 doubly transfected HeLa cells (Fig. 5 C and D and Fig. S7 D and E). The mutant showed a significant increase in basal Tie2 phosphorylation in both singly and doubly transfected cells. Interestingly, the Fn2 mutant could be stimulated with Comp-Ang1 in the presence of Tie1 but not in the absence of Tie1. Alignment of Ang1-bound homodimers of the Tie2 ECDs creates a model for Tie2 clustering (Fig. 5E). Notably, multiple Tie2 homodimers can interact without steric clashes, and the model brings the neighboring LBDs close to each other.

Crystallization and Structure Determination.

Human Tie2 Fn1–3 (residues 443–735) and human Tie1 Fn3 (residues 641–738) were expressed in insect cells, and the purified proteins were crystallized for structure determination. Full details of protein purification and X-ray and SAXS data collection and analysis are described in SI Methods.

Cell Culture and Phosphorylation of Tie1 and Tie2.

WT and mutant Tie1 and Tie2 were cloned into pMXs retroviral expression vector that was used to transduce HeLa cells as described by Saharinen et al. (10). Full details of cell culture and analysis are described in SI Methods.

Protein Expression and Purification.

For crystallization, human Tie2 Fn1–3 (residues 443–735) and human Tie1 Fn3 (residues 641–738) were cloned into the pFastBac (Invitrogen) baculovirus expression vector with a C-terminal Factor Xa cleavage site and a His-tag. For analytical gel filtration and SAXS analysis, human Tie2 ECD (residues 1–742), Fn2–3 (residues 542–742), and Fn3 domain (residues 635–742) were similarly cloned into the pFastBac (Invitrogen) baculovirus expression vector. For expression analysis, human Tie2 double-mutants Val685Tyr/Val687Tyr and Gln588Arg/Val612Arg were cloned into the Fn1–3 domains. Recombinant baculovirus was produced in Sf9 insect cells in serum-free Insect-Express (Lonza) medium supplemented with 50 μg/mL gentamycin (Sigma) at 26 °C. For protein expression, Tn5 insect cells were infected with the corresponding recombinant baculovirus at high multiplicity, and the supernatant was harvested by centrifugation at 3 d postinfection. The His-tagged proteins were extracted by Ni²⁺-charged chelating Sepharose (GE Healthcare), the resin was washed in PBS containing 20 mM imidazole, and the proteins were eluted with 0.4 M imidazole in PBS. Finally, followed by the His-tag removal, the proteins were purified by gel filtration on a Superdex 200 (GE Healthcare) column in 10 mM Hepes-buffered saline (HBS) (Hepes, 0.15 M NaCl at pH 7.5).

Crystallization and Structure Determination.

For crystallization, Tie2 Fn1–3 and Tie1 Fn3 proteins were concentrated to 10 mg/mL in HBS buffer supplemented with 0.1% (vol/vol) P8340 protease inhibitor mixture (Sigma) and 0.01% (wt/vol) NaN₃. Crystallization conditions were screened using the sitting-drop vapor-diffusion technique. Monoclinic Tie2 Fn1–3 crystals in space groups C2 grew in a few days at room temperature over a reservoir solution of 0.1 M Tris buffer at pH 7.0–8.5 and 14–20% PEG 3350 (wt/vol). Within the initial crystals in space group C2, in 5–6 wk grew a single larger crystal that processed in space group P2₁. Tie1 Fn3 crystallized in space group P6₅ in 0.1 M Tris buffer at pH 8.0–9.0 and 18–24% PEG 8000 (wt/vol). The drops were prepared by manually mixing 1–2 μL of the reservoir solution and 1–2 μL of the protein solution. The Tie2 Fn1–3 crystals in space groups P2₁ (a = 78.5 Å, b = 48.5 Å, c = 80.7 Å, and β = 111.1) and C2 (a = 134.2 Å, b = 104.7 Å, c = 79.5 Å, and β = 121.4) contain very similar homodimers per asymmetric units. However, the two protein chains in the P2₁ crystal lacked the Fn1 domains, whereas in the C2 crystals one chain was intact but the other Fn1 domain was either disordered or removed by proteolysis. Tie1 Fn3 crystals (a, b = 54.1 Å, c = 107.3 Å) contain a strand-swapped homodimer in the asymmetric unit. For heavy-atom derivative data collection, the Tie1 Fn3 and Tie2 Fn1–3 crystals in C2 were soaked in 1% (vol/vol) of saturated solution of methylmercury acetate at pH 8.0 for 16 h or in 0.1 mM potassium tetrachloroplatinate for 1 h. For data collection at 100 K, the crystals were frozen in liquid nitrogen in the reservoir solution supplemented with 20% glycerol.

Complete datasets to 2.6 Å and 2.9 Å resolution were collected from native P2₁ and C2 crystals of the Tie2 Fn-like domains, respectively, at the European Synchrotron Radiation Facility (ESRF) beamlines in Grenoble, France (Table S1). Data were processed with XDS (52) and the CCP4 suite of programs. The Tie2 Fn1–3 structure in C2 was solved by MIRAS-phasing using autoSHARP (53). Automated model building in PHENIX (54) was used to generate an initial model and improve the phases. The model was completed by iterative refinement in Refmac and model building in Coot. A subset of 5% of the diffraction data was omitted from refinement for calculating the free R factor (R_free). The Tie2 Fn2–3 structure in P2₁ was solved by molecular replacement using the Fn2 and Fn3 domains of the C2 structure as search models. The final model in P2₁ contains residues 543–568 and 571–734 in chain A and residues 542–567 and 570–734 in chain B, N-linked glucosamine moieties in Asn560 and Asn596 in Fn2, Asn649 in Fn3, and 62 solvent molecules. Of the residues, 94.4% are within the most-favored region of the Ramachandran plot. The final model in C2 contains residues 443–524 and 528–735 in chain A and residues 540–571 and 573–731 in chain B, N-linked glucosamine moieties also in Asn464 in Fn1, and 63 solvent molecules. In addition, chain A has N-terminal aspartic acid and proline, numbered as 441 and 442, as cloning artifacts. Of the residues, 96.2% are within the most-favored region of the Ramachandran plot. The homodimers of Tie2 Fn-like domains were superimposed in Coot, and the rmsd value for the comparison of 376 Cα-atoms was calculated with LSQKAB in the CCP4 suite. Wilson B-factors were calculated with the Sfcheck program in the CCP4 suite.

Data of the Tie1 Fn3 crystals (Table S1) were collected at the ESRF beamline ID-30B and processed with XDS (52). Heavy-atom phasing was hampered by nearly perfect twinning (k,h,−l), and the structure was solved using the Tie2 Fn3 domain as a search model in molecular replacement. The final model contains residues 643–738 in chain A and residues 644–737 in chain B. Residues 686–688 in chain A and residues 684–689 in chain B are disordered. Ninety percent of the residues are within the most-favored region of the Ramachandran plot. All molecular graphics figures were generated with PyMOL (W. Delano; www.pymol.org/).

SAXS Data Collection and Analysis.

The protein samples for SAXS were concentrated by using Amicon Ultra-4 centrifugal filter units (EMD Millipore) and were frozen for storage at −80 °C. The samples were purified by size-exclusion chromatography column in 20 mM Hepes (pH 7.5), 0.15 M NaCl, and 5% glycerol, and the running buffer was used as a solvent blank for SAXS. SAXS data were collected on beamline BM29 at the ESRF (Grenoble, France). Each sample had 10 exposures taken at a sample-to-detector distance of 2.867 m and at 12.5 keV, covering a momentum transfer q-range of 0.0025–0.5/Å. Samples were passed through the beam continuously using a flowthrough capillary cell to minimize radiation damage. The 10 consecutive exposures were compared for change in scattering intensity caused by radiation damage. Radially averaged scattering data were buffer subtracted and analyzed by using ATSAS (2.7.1) software (55). Scattering curves were displayed as the scattering intensity [I(q)] as a function of momentum transfer q = (4πsinθ)/λ, where λ is the wavelength of the incident X-ray beam, and θ is half the angle between the incident and scattering radiation. SAXS data were obtained for four different concentrations of Tie1 Fn3 protein and were checked for aggregation by examination of the linearity in the Guinier region in log I(q) versus log (q) plots, a diagnostic of sample quality.

The radius of gyration, R_g, was calculated from the scattering curve by using the indirect Fourier transform program AUTOGNOM, which additionally provides an estimate of the maximum particle dimension (D_max) from the distance distribution function P(r). AUTOGNOM also was used to assign the q range (q_max <8/R_g) for ab initio shape reconstructions by DAMMIF, which uses simulated annealing methods to build low-resolution protein models from densely packed beads inside a search volume. The final model was obtained by aligning 10 independent models to produce an averaged, volume-filtered model. Experimental scattering curves were compared with theoretical scattering curves calculated by the program CRYSOL (for q_max <0.5/Å) from the crystal structures of Tie1 Fn3 and Tie2 Fn2–3.

Multiangle Laser Light Scattering.

SEC-MALLS measurements were run at 0.5 mL/min over an S-200 Superdex 10/300 column (GE Healthcare) in PBS with a Schimadzu HPLC system and MiniDAWN TREOS light-scattering detector and Optilab rEX refractive index detector (Wyatt Technology Corporation). Data were analyzed with ASTRA 6 software (Wyatt Technology Corporation).

Cell Culture and Retroviral Generation.

HeLa cells were maintained in DMEM (Lonza) supplemented with 2 mM l-glutamine, penicillin (100 U/mL), streptomycin (100 μg/mL), and 10% FBS. The retrovirus packaging cell line 293-GPG VSV-G51 was maintained in DMEM (Lonza) with 4.5 g/L glucose and supplemented with 10% FBS, 1% ​l-glutamine, 0.2% penicillin, 0.2% streptomycin, 0.2% puromycin, 0.6% neomycin, and 1 μg/mL tetracycline. The 293-GPG VSV-G51 cells were transformed with pMXs retroviral expression vectors with Fugene 6 (Roche) in DMEM supplemented with 20 mM Hepes, pH 7.5. Equal amounts of WT Tie1 or Tie2 or mutant Tie receptor retrovirus containing supernatants were used to transduce HeLa cells. Tie2 double mutants Val685Tyr/Val687Tyr and Gln588Arg/Val612Arg were cloned into the Myc-tagged, full-length human Tie2, and the Tie1 triple mutant V681Y/L691Y/I693Y and the quintuple mutant with additional D689R/D696R mutations were cloned into the Strep-tagged, full-length human Tie1. GFP-tagged human Tie1 was used for immunofluorescence staining of Tie1/Tie2 double-transfected cells.

Phosphorylation of Tie1 and Tie2.

For Ang1 stimulations, cells were starved overnight in 1% FBS containing medium, followed by 1 h stimulation with 100 ng/mL Comp-Ang1. The cells were lysed as described previously in 1% Triton X-100 containing PLCLB (50 mM Hepes, 150 mM NaCl, 10% glycerol, 1% Triton X-100, 1.5 mM MgCl₂, 1 mM EGTA, 10 mM Na₄P₂O₇, 100 mM NaF) lysis buffer (10, 21) followed by measurement of protein concentration and immunoprecipitation using anti-Tie2 (AF313; R&D Systems) or anti-Tie1 (AF619; R&D Systems) antibodies and protein G-Sepharose (GE Health Sciences Ab) at +4 °C. The protein concentration of the lysates was measured using the Pierce BCA Protein Assay kit (Thermo Scientific). The immunocomplexes were separated in SDS/PAGE (Bio-Rad Laboratories) and were transferred to PVDF membrane (Millipore) using the Trans-Blot Turbo Transfer System (Bio-Rad Laboratories). Membranes were used for immunoblotting with anti–phospho-Tie2 (pY-992, AF2720; R&D Systems), anti-Tie2 (AF313; R&D Systems), or anti-Tie1 (AF619; R&D Systems) as primary antibodies and HRP-conjugated secondary antibodies (Dako) or biotinylated secondary antibodies (Dako) followed by ECL detection with the SuperSignal West Pico Chemiluminescent Substrate or SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Scientific). Western blots were imaged with Odyssey FC (LI-COR). For reprobing, the membranes were stripped using the ReBlot Plus Strong Antibody stripping buffer (Millipore). Western blots were quantified using ImageJ. The pY/Tie receptor ratios were normalized by setting the pY/Tie2 ratios of unstimulated Tie2-WT/Tie1-WT samples to 1 in Fig. 2D and the pY/Tie2 ratios of Comp-Ang1–stimulated Tie1-WT/Tie2-WT samples to 1 in Figs. 4D and 5D.

Immunofluorescence Imaging.

For immunofluorescence staining of HeLa cells, the cells were fixed with 4% paraformaldehyde for 10 min, permeabilized with 0.1% Triton-X in PBS for 5 min, blocked with 1% BSA-PBS (Biotop), stained using rabbit anti–phospho-Tie2 (AF2720; R&D Systems) as primary antibodies for 30 min at room temperature, washed, and stained with appropriate Alexa-594–conjugated antibody (Life Technologies). The samples were mounted with Vectashield mounting medium containing DAPI (H-1200; Vector Labs). Images were captured with a Hamamatsu digital camera (ORCA-Flash4.0 LT) connected to a Zeiss Axio Imager 2 upright epifluorescence microscope.

Statistical Analysis.

Student’s t test (two tailed) was used for pairwise comparisons. A P value <0.05 was considered statistically significant.